Hello
I am running a binary MLM and I was wondering whether:
a. I check the assumption of normality of residual as I would do for a linear MLM (via Normal probability plots) .
b. I can check the assumtion of linearity of the logit in MLwiN.
c. I can check for outliers in MLwiN.
Thank you
Binary MLM assumptions
Re: Binary MLM assumptions
Hi Stasis,
At higher levels in the model you are assuming normality of residuals so you can test normality here as per normal models. Not much point at looking at level 1 residuals here as they often form 2 bands - one for the 1s and one for the 0s!
I guess the concept of outliers is difficult when all data are 0 or 1 but you could have an outlying level 2 unit which had more 0s or 1s than expected from the normal distribution of residuals - this would show up in your normality plot.
As for the linearity question - not sure I fully follow but you could test linearity of continuous predictors by adding higher powers of these in the model and seeing if they are significant.
Hope this helps,
Best wishes,
Bill.
At higher levels in the model you are assuming normality of residuals so you can test normality here as per normal models. Not much point at looking at level 1 residuals here as they often form 2 bands - one for the 1s and one for the 0s!
I guess the concept of outliers is difficult when all data are 0 or 1 but you could have an outlying level 2 unit which had more 0s or 1s than expected from the normal distribution of residuals - this would show up in your normality plot.
As for the linearity question - not sure I fully follow but you could test linearity of continuous predictors by adding higher powers of these in the model and seeing if they are significant.
Hope this helps,
Best wishes,
Bill.